David Leibner: Just wanna thank you for the opportunity to present a little bit of what 
I've been doing, and highlight a few of the areas that we've been working on. In particular, 
a little bit of introduction; my name is David Leibner, actually I'm a practicing medical 
oncologist. I also have a joint appointment through the Department of Biomedical 
Informatics, and my work is focused primarily in the area of cancer analytics.

So, big data has become a bit of a buzzword. But with the recent large-scale projects, both 
in the United States and Europe, we are generating an enormous amount of data, both molecular 
as well as epigenetic information on cancer. And we're no longer at a loss for the volume of 
information.

But currently, the question is, how do we best make sense of this? So there are several 
large-scale projects, as I mentioned, one of them here in the United States is the cancer 
genome atlas. Definitely, I said there's been a number of ways that cancers are now 
characterized: mutation numbers, copy number abnormalities, gene expression, DNA 
methylation, microRNA.

And all of these are now being combined in ways to try to make sense of how different 
types of cancers behave differently, and what are the potential targets for therapeutic 
intervention. 

[Text on screen 
"Big Data" Has Arrived in Cancer Analytics]

[Graphs displaying the 12 tumor types and Omics characteristics] 

So this has really been my focus of work, both in the analytic side, the development of 
tools for analyzing this data, and with the goal of bringing this into a way that's 
applicable to the clinical arena for patients with cancer.

I'm just gonna highlight two areas of active interest right now. The first one is in the 
realm of tumor-stromal interactions. We know that cancer is not just a homogeneous pool 
of cancer cells; there definitely are stromal interactions. So, stromal, referring to the 
immune cells, the normal connective tissue that coexists with the malignant cancer clonal 
cancer cells, and understanding how this interplay between the tumor and the stroma can be 
related to patient outcomes, as well as potentially the response to therapeutics. And how 
we actually approach the problem of big data, how we utilize some of the datasets
that are available publicly, to understand these issues and to generate new hypotheses, and 
then, subsequently, test these hypotheses in the clinical arena. 

And then the second area that I'm going to focus on a little bit is how we actually integrate 
all of this information in a way that makes use of the information flow within cancer cells.

This touches a bit on systems biology, but our current understanding of how information is 
transmitted through cancer cells between proteins can be applied to the understanding of 
why certain mutations and certain pathways are more or less important in different cancer 
subtypes, and developing tools to help us understand those questions.

[Text on screen 
Major Areas of Interest 
Focus is on the developing of analytical models to handle "big data" from cancer studies 
to better understand: 
-Tumor-stromal interactions, particularly: 
Impact on patient outcomes, Impact on response to therapeutics
-Cancer genetics, particularly: 
Using information theory to understand why there are differences in driver mutations in 
different cancer types, and predicting pathways mechanisms of resistance.] 

So, for the first hypothesis that I'm gonna focus on, or the potential project, is: can 
computational approaches, or I say, described as computational microdissection, looking 
at expression of expression profiling of an individual tumor, be used in a way that allows 
us to predict responses to cancer therapies that are targeted at the tumor stroma?

In particular, if we're looking at anti-angiogenic drugs or immunotherapies, and developing 
the tools that allow us to ask and to pose those questions, as well as to potentially answer 
those questions. So, my lab has worked on developing those tools. We do have applications 
that actually run primarily using MATLAB computer code.

But there certainly are other computer and informatics-based applications that can take 
advantage of any of the computing resources, be it benchtop or in the supercomputer. And 
then kind of use these to look at clinical data that's collected either in public data sets, 
such as Cancer Genome Atlas, or as we move forward in patients who are here at Ohio State.

[Text on screen 
Hypotheses/Projects for Trainees
1. Computational microdissection can be used to predict response to cancer therapies that 
are targeted at tumor stroma (anti-angiogenic drugs, immunotherapies).] 

[Graph displaying the relationship between overall survival, recurrence risk, and resistance 
to treatment] 

So, as I said earlier, the problem of tumor heterogeneity: tumors do not simply consist of 
a pure collection of tumor cells. There's an addition to tumor cells; you have immune cells, 
you have fibroblasts or other mesenchymal-type cells that are actually infiltrating the 
tumor, you have neovascularization. 

[Text on screen 
Appreciating the "Problem" of Tumor Heterogeneity] 

[Diagram showing the complexity of a tumor, illustrating types of cells, signaling pathways, 
and the tumor microenvironment.] 

And so the question is, when we get a biopsy, people often think of it conceptually as a pure 
representation of the cancer, looking for potential biomarkers of response.

But, if we actually understand things better, it's actually each cell type may have different 
molecular predictors of response. And so, what we've been doing is taking these tissue 
samples with mixed cellularity, breaking them down, looking at individual genes, looking at 
the expression using computational approaches, and then breaking it down into what appears to 
be different cell populations in each tumor type.

And we can actually look at gene expression even without actually dissecting them out 
physically. We can look at gene expression on a cell-by-cell basis, based on the population 
of cells. And so these tools allow us to, for example, predict stromal composition in 
individual tumor biopsies, and this is from a series of 44 patients with metastatic 
melanoma.

[Diagrams showing a computational analysis of mRNA profiles in heterogeneous tissues] 

We're able to show that there are fairly significant immune infiltrates in many of these 
patients. There's varying degrees of actual malignant clones, at least by characteristic 
gene expression profile in each of these populations. And we can use this information to 
really look and say, well, for each of these cell types, so B cells are an important part 
of the immune system, the adaptive immune system, which produces antibodies, T cells, and 
NK cells, we think as being part of the direct cytotoxic, the direct tumor killing parts 
of the immune system.

[Bar chart showing the predicted stromal composition in a series of melanoma biopsies, 
specifically studying the relationship between samples and the estimated RNA fraction] 

Then we actually look and break down and say, do any of these factors correlate to patient 
outcomes? And we have preliminary data which shows that in patients who have high 
concentrations of T cells within their tumor biopsies, those patients do significantly
better than patients who have fewer T cells.

And we can repeat this, this is a very not fine-grained, not very sophisticated analysis, 
but there are more sophisticated models that you can build based on the types of immune 
cells that are present. And so, the question is how we can subsequently translate this, 
not merely to patient outcomes, but actually looking as earlier stromal-directed 
therapies.

[Graph displaying a survival analysis of stage III and IV melanoma, specifically studying 
the time in days relationshipn to overall survival from time of metastatic disease] 

The second project that is currently in development that I wanted to talk about today is 
how we use the idea of information flow within protein networks, protein-protein 
interaction, to try to understand why different pathways are activated, and more important 
in different types of cancers. Why are certain genes more likely to be mutated in one 
cancer type than the other?

To understand those mutation pressures, and by understanding those mutation pressures, 
identify potential genes that may be additionally, that may not originally be thought of 
as a target, but additionally, genes that may be a mechanism of escaping a targeted 
therapy. 

[Text on screen 
Hypotheses/Projects for Trainees 
2. Information flow in annotated protein interaction networks (e.., STRING-DB) can be 
integrated with RNA expression data (e.g., TCGA) to: 
-Understand patterns of mutations in different cancer types 
-Understand mutation pressures, and 
-Propose potential pathways of resistance to targeted therapies.] 

This is obviously a very simplistic, but helps us conceptually understand, for example, 
in cancer, one pathway that's involved.

And we can see that there's a very systematic interaction where information is passed from, 
for example, growth factors to a receptor at the self surface, and this is subsequently 
passed. This is the Ras/Raf pathway, ultimately leading to expression of genes involved 
in proliferation, differentiation, and survival. And so in melanoma, we see that 
mutations will often occur within the Ras gene, within the Raf gene, and then less so 
often within the MAP kinase genes.

And so the question becomes, how are Ras and Raf similar? How is information carried by those genes within these pathways? And how can we use that to understand why these are such important linchpins? And by looking at where they fall within the information flow within the cell, we can identify similarities and differences.

[Flow chart displaying Protein interaction networks control information flow within 
cells] 

We can also look at not only information flow, but as I said, other groups have already 
started to look at how different types of mutations tend to cluster within certain
pathways. So the question is, as we pull together all this information from individual, 
from these large public databases to understand for different cancer types, why 
clusters occur where they do.

[Cluster chart displaying mutation patterns in different cancers based on similar 
function] 

This is a lot messier, but this is a little bit more realistic than the original 
simplistic information flow. But we can see that lots of genes and lots of proteins 
are connected. Each dot here represents an individual gene or protein, each line 
represents a connection between one gene or one and another gene, at least that has 
been documented thus far.

[Cluster chart labeled "Modeling Information Flow in Interaction Networks Allows us 
to Identify Similar Genes] 

And looking how information flows and how epigenetic data that we have available, 
how it modulates information flow, and why things differ. So just to sum things up, 
two major areas of focus are really looking at that computational microdissection, 
looking at tumor stromal, looking at how that impacts patient outcomes.

Expanding this to additional projects here at Ohio State, as well as using public 
database to refine the existing methodology, as well as looking at information flow 
and protein interaction networks. How can we integrate this with additional data 
that we have, to better understand the mutation patterns that we see in different 
cancer subtypes.

And use that to help us guide decisions based on to predict therapeutic response 
to targeted agents, as well as resistance to targeted agents. So, thank you very 
much.

[Text on screen 
Hypotheses/Projects for Trainees 
1. Computational microdissection can be used to predict response to cancer therapies 
that are targeted at tumor stroma (anti-angiogenic drugs, immunotherapies). 
2. Information flow in annotated protein interaction networks (e.g., STRING-DB) 
can be integrated with RNA expression data (e.g., TCGA) ro: 
-understand patterns of mutations in different cancer types 
-Understand mutation pressures, and 
-Propose potential pathways of resistance to targeted therapies.]